Active Aerial Search in Forests: Multi-UAV Next-Best-View Planning with Prior Scene Data

Abstract

Autonomous aerial search in dense forest environments remains challenging due to severe occlusion, limited sensor visibility, and uncertainty in prior scene models. This work presents a multi-Unmanned Aerial Vehicle (UAV) framework for active search that integrates visibility-driven Next Best View (NBV) planning with centralized camera pose allocation informed by prior 3D environmental data. The system adaptively selects camera poses that maximize new visibility in cluttered regions while distributing search effort efficiently across multiple agents. Field experiments in vegetated forests, searching for occluded objects, demonstrate that coordinated NBV planning substantially improves detection performance. Notably, the generated oblique camera poses enabled the detection of all hidden objects that otherwise are hidden for nadir views. These findings highlight the effectiveness of combining multi-agent coordination with visibility-aware NBV strategies, offering a scalable and robust approach for large-scale aerial search and challenging Search and Rescue (SaR) scenarios.